International Business Machines Corporation

Karthik Nandakumar

Nalini Kanta Ratha

Sharathchandra UmapathiRao Pankanti

Shai Halevi

Roman Vaculin

Karthikeyan Shanmugam

Kanthi K Sarpatwar

Sebastien Blandin

Sharathchandra Umapathirao Pankanti

Chaitanya Kumar

Laura Wynter

Yuan Yuan

James Thomas Rayfield

Yanyan Hu

Shelby Solomon Darnell

Shengjiao Cao

ML model(s) are created and trained using training data from user(s) to create corresponding trained ML model(s). The training data is in FHE domains, each FHE domain corresponding to an individual one of the user(s). The trained machine learning model(s) are run to perform inferencing using other data from at least one of the user(s). The running of the ML model(s) determines results. The other data is in a corresponding FHE domain of the at least one user. Using at least the results, it is determined which of the following issues is true: the results comprise objectionable material, or at least one of the trained ML model(s) performs prohibited release of information. One or more actions are taken to take to address the issue determined to be true. Methods, apparatus, and computer program product are disclosed.

Searching over encrypted model and encrypted data using secure single-and multi-party learning based on encrypted data

Securely re-encrypting homomorphically encrypted data by receiving fully homomorphically encrypted (FHE) information from a client device, training a machine learning model using the FHE information, yielding FHE ciphertexts, applying a first transform to the FHE ciphertexts, yielding obfuscated FHE ciphertexts, sending the obfuscated FHE ciphertexts to a secure device, receiving a re-encrypted version of the obfuscated FHE ciphertexts from the secure device, applying a second transform to the re-encrypted version of the obfuscated FHE ciphertexts yielding de-obfuscated re-encrypted FHE ciphertexts, determining FHE ML model parameters according to the de-obfuscated re-encrypted ciphertexts, and sending the FHE ML model parameters to the client device.

Secure re-encryption of homomorphically encrypted data

A framework is provided in which a querying agency can request (via a query entity) encrypted data through a service provider from a data owning agency that stores encrypted data. The framework uses homomorphic encryption. The data may be gallery entities, and each of the elements in the framework operate on doubly-encrypted information. The service provider compares a representation of an encrypted query entity from the querying agency and representations of encrypted gallery entities from the data owning agency, resulting in doubly-encrypted values of a metric between corresponding compared representations. The querying agency gets result(s), based on the metric, which indicate whether it is probable the service provider has data similar to or the same as query data in the query entity. The elements have to perform communication in order for the querying agency or the data owning agency to get cleartext information corresponding to the query entity.

Secure matching and identification of patterns

A method, apparatus and computer program product for homomorphic inference on a decision tree (DT) model. In lieu of HE-based inferencing on the decision tree, the inferencing instead is performed on a neural network (NN), which acts as a surrogate. To this end, the neural network is trained to learn DT decision boundaries, preferably without using the original DT model data training points. During training, a random data set is applied to the DT, and expected outputs are recorded. This random data set and the expected outputs are then used to train the neural network such that the outputs of the neural network match the outputs expected from applying the original data set to the DT. Preferably, the neural network has low depth, just a few layers. HE-based inferencing on the decision tree is done using HE inferencing on the shallow neural network. The latter is computationally-efficient and is carried without the need for bootstrapping.

Decision tree-based inference on homomorphically-encrypted data without bootstrapping

A data intersection is assessed of data to be used between at least two parties. The data is to be used in an artificial intelligence (AI) application. Evaluation is performed of set of instructions required for the AI application, where the evaluation creates a modified set of instructions where operands are symbolically associated with corresponding privacy levels. Using the assessed data intersection and the modified set of instructions, a mapping is created from the data to operands with associated privacy metrics. The mapping treats overlapping data from the assessed data intersection differently from data that is not overlapping to improve privacy relative to without the mapping. The AI application is executed using the data to produce at least one parameter of the AI application. The at least one parameter is output for use for a trained version of the AI application. Apparatus, methods, and computer program products are described.

Collaborative AI on transactional data with privacy guarantees

This disclosure provides a method, apparatus and computer program product to create a full homomorphic encryption (FHE)-friendly machine learning model. The approach herein leverages a knowledge distillation framework wherein the FHE-friendly (student) ML model closely mimics the predictions of a more complex (teacher) model, wherein the teacher model is one that, relative to the student model, is more complex and that is pre-trained on large datasets. In the approach herein, the distillation framework uses the more complex teacher model to facilitate training of the FHE-friendly model, but using synthetically-generated training data in lieu of the original datasets used to train the teacher.

Depth-constrained knowledge distillation for inference on encrypted data

A technique for computationally-efficient privacy-preserving homomorphic inferencing against a decision tree. Inferencing is carried out by a server against encrypted data points provided by a client. Fully homomorphic computation is enabled with respect to the decision tree by intelligently configuring the tree and the real number-valued features that are applied to the tree. To that end, and to the extent the decision tree is unbalanced, the server first balances the tree. A cryptographic packing scheme is then applied to the balanced decision tree and, in particular, to one or more entries in at least one of: an encrypted feature set, and a threshold data set, that are to be used during the decision tree evaluation process. Upon receipt of an encrypted data point, homomorphic inferencing on the configured decision tree is performed using a highly-accurate approximation comparator, which implements a 'soft' membership recursive computation on real numbers, all in an oblivious manner.

Privacy-enhanced decision tree-based inference on homomorphically-encrypted data

An AI model is trained by determining insights for a sequence of computations used in the AI model. The sequence is applied to encrypted data and label pair(s), wherein computational details of each of the computations are defined. Information may also be committed for selected ones of the sequence of computations into a distributed database. The committed information may include computational details used in processing performed for the selected computations, and the distributed database may have a property that the committed information for each selected computation is linked with a verifiable signature of integrity with a previously committed computation in the sequence. Indication is received from an end-user computer system of selected computation(s). Computational details of the indicated selected computation(s) are sent toward the end-user computer system for use by the end-user computer system for verifying the indicated selected computation(s). The end-user computer system can verify the selected computation(s).

Encrypted data model verification

Respective sets of homomorphically encrypted training data are received from multiple users, each encrypted by a key of a respective user. The respective sets are provided to a combined machine learning model to determine corresponding locally learned outputs, each in an FHE domain of one of the users. Conversion is coordinated of the locally learned outputs in the FHE domains into an MFHE domain, where each converted locally learned output is encrypted by all of the users. The converted locally learned outputs are aggregated into a converted composite output in the MFHE domain. A conversion is coordinated of the converted composite output in the MFHE domain into the FHE domains of the corresponding users, where each converted decrypted composite output is encrypted by only a respective one of the users. The combined machine learning model is updated based on the converted composite outputs. The model may be used for inferencing.

Secure multi-party learning and inferring insights based on encrypted data

An example operation may include one or more of obtaining a first biometric sample of a user from a user device. extracting, by an issuing node of a permissioned blockchain network, a biometric template from the first biometric sample, encrypting the biometric template, distributing an issuetoken proposal comprising the encrypted biometric template to the blockchain network, and generating and distributing a biometric token to the user device. In response to the user indicating to the user device to redeem the biometric token, the method includes one or more of presenting, by the user device, the biometric token to a verifying node of the blockchain network, validating, by the verifying node, the biometric token, receiving, by the verifying node, a second biometric sample from the user device, distributing a redeemtoken proposal to the blockchain network, committing a transaction corresponding to the biometric token, to the blockchain network, and invalidating the biometric token.

Biometric token for blockchain

A computer system accesses and processes regulatory requirements for data item(s) in a private manner. Both the data item(s) and the regulatory requirements are accessed and processed privately. The computer system creates an orchestration strategy satisfying the regulatory requirements. The orchestration strategy includes recommendation(s) associating the data item(s) with process(es). The computer system outputs indications of the orchestration strategy to be used to implement regulatory compliance for processing of the data item(s) by associated ones of the process(es). The computer system may be implemented as a portion of a cloud environment, and compliance may be offered as a service for cases where data usage by an application (implementing the process(es)) does not address compliance with the regulatory requirements, but following the orchestration strategy ensures use of the application on the data item(s) will comply with the regulatory requirements.

Regulatory compliance for applications applicable to providing a service for regulatory compliance on a cloud

An example operation may include one or more of initiating, by a file verification device, verification of a source file or a redacted source file, executing one of a smart contract or chaincode to verify the chameleon hash signature and the auxiliary data hash signature, and providing a notification whether the verification was successful or unsuccessful. In response to initiating verification of the source file, the method further includes the file verification device receiving stored source file segments and stored auxiliary data segments, generating a chameleon hash signature, and generating an auxiliary data hash signature. In response to initiating verification of the redacted source file, the method further includes receiving stored redacted file segments, stored auxiliary data segments, and stored modified auxiliary data, generating a chameleon hash signature, and generating an auxiliary data hash signature.

File verification database system

An example operation may include one or more of receiving, at an endorser node, a request message from a client system which comprises data to be stored on a blockchain, determining whether to endorse the data via invocation of chaincode which receives the data as input and executes the data against a current state of the blockchain, in response to a determination to endorse the data, generating a response message including a result of the execution and signing the response message based on a traceable blinded ring signature associated with the endorser node, and transmitting the generated response message that has been signed with the traceable blinded key ring to the client system.

Growing community of inventors

Singapore, Singapore

Karthik Nandakumar